L2 Morpheme Acquisition Order Across L1 Language Families
Elicited imitation (EI) is an oral testing methodology designed to test a language learner’s understanding of specific grammatical structures (Erlam, 2006). In administering EI tests, a subject listens to a string of words and then is asked to reproduce that string of words. Theory asserts that each subject forms a cognitive representation based on the grammatical structures understood in the string of words presented, and then produces a string of words according to that representation (Bley-Vroman & Chaudron, 1994). A person’s ability to reproduce a sentence is connected with that person’s understanding of the structure of the language. EI has provided helpful insight into the process of language acquisition and language testing.
In research on the learning of English as a second language (ESL), learners’ first language (L1) backgrounds are an important consideration. Similarity between English and a learner’s L1 differs greatly depending on the L1, and there is still much research to be done on specific grammatical structures in English that are most problematic for speakers of different L1s. Odlin (2003) explains that similarities and differences between a person’s L1 and second language (L2) affect the acquisition of the L2. There is research on general grammatical structures that are problematic for specific L1s, such as plurality and negation (Romaine, 2003), but precise morphological structures of English most problematic for different L1 language families have not been compared and contrasted through one test.
In this presentation we report on a pilot study designed to determine how subjects’ L1 affects the natural order of L2 morpheme acquisition in English. DeKeyser and Goldschneider (2005) cite 6 morphemes in English—including progressive, regular past tense, and plural—and conduct a large scale survey of the research indicating the order of acquisition of the same 6 morphemes. While DeKeyser and Goldschneider’s survey is extensive, the studies they survey are not entirely in agreement, and generalizations are made to the entire population of English L2 learners; the authors do not examine differences between morpheme acquisition for varying L1s.
In our study, using EI results from approximately five hundred ESL students, we investigate morpheme acquisition order according to L1. We divide subjects into groups by L1 using the ten primary languages and the four main language families in our subject pool and conduct multiple analyses of variance to determine similarities and differences between these groups in terms of accuracy levels for EI items involving each of the morphemes highlighted in the DeKeyser and Goldschneider study. We will report these analyses and discuss possible implications regarding L1 background and L2 acquisition order.
[Jeremiah McGhee, Malena Weitze and Dan Dewey]
Improving automated oral testing: identifying features and enhancing speech recognition.
Elicited imitation (EI) is an oral testing methodology that has enjoyed some resurgence in recent years for adult L2 learners (Erlam, 2006). Due to the closed-world assumption inherent in the testing procedure (the responses are known a priori), the results are particularly well adapted to automatic analysis techniques including speech recognition.
We discuss the multiple EI tests developed and administered that specifically target grammatical features found to be problematic for EFL learners. In addition, we discuss how the sentences in these tests were hand-crafted to explicitly test learner’s knowledge of these grammatical features, and give relevant examples of test sentences and sample protocols. We explain how the tests were scored and the comparative results between different iterations of our EI tests.
Currently, we have over a million EI syllables graded according to accuracy of reproduction which have been compiled into a comprehensive corpus. We have thus been able to extend our previous analyses of EI items in terms of relative complexity and correlation with other oral testing modalities. Many of the items have been scored by two annotators, and analyzing the discrepancies leads to interesting observations about the data itself, the annotators, and the scoring methodology.
We are also evaluating to what extent the scoring procedure can be enhanced by using attributes external to the EI utterances and annotations themselves, for example taking into consideration other information about the test subjects (e.g. their first language, age, reading proficiency scores, fluency measures, etc.). This entails questions, empirical and theoretical, about how to situate the EI test results in a larger, holistic evaluation context. We discuss which statistical and machine learning techniques have proven most useful to date in our work.
[The PSST Research Group]